I'm working on a regression. The dependent variable is binary, and most of my independent variables are continuous, except one. For about 40% of the "age" variable, the data only says "> 5".

How in the heck do I treat that? Replace with the mean? Treat age as a dummy variable (really don't want to do this)? Consider it missing (ditto)?

  • 1
    $\begingroup$ Can you split on the discrete variable? How many levels is it? How many observations do you have? $\endgroup$ – EngrStudent Dec 20 '16 at 15:42
  • 3
    $\begingroup$ Wait... so from 6 through 96 they are all coded as ">5"? $\endgroup$ – Penguin_Knight Dec 20 '16 at 15:49
  • 1
    $\begingroup$ Can you make a new variable that splits on over 5 versus 5 and under? This would give you two levels. You might try to predict age given the other measures, and then use that. Can you predict age given the other variables, then split model on the predicted age? Does this work for all ages? $\endgroup$ – EngrStudent Dec 20 '16 at 15:53
  • 2
    $\begingroup$ You could set it to missing and then do multiple imputation so it will end up being replaced with best guess values for some meaning of best guess. $\endgroup$ – mdewey Dec 20 '16 at 15:54
  • 2
    $\begingroup$ I'd also suggest checking if the data set is compiled/entered correctly. I had situations in which this ">5" could be one of the answer for the next question but for some reason an accidental skip happened and errors like this happened. $\endgroup$ – Penguin_Knight Dec 20 '16 at 16:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.